Role of machine learning techniques in classification of peripheral pulse morphology

Abstract

Recording of peripheral pulse serves as a very important and essential non-invasive tool used widely by doctors for the diagnosis of various diseases. The morphology of pulse is seen to vary as a function of time in a given individual and also from individual to individual. There are many variations in morphological patterns of peripheral pulse in different disease conditions, which lead to difficulty in accurate diagnosis. The peripheral pulse waveforms are extracted from radial arteries as time series data using a peripheral pulse analyzer which is designed on the principle of impedance plethysmography. It was first introduced by Nyober in the mid-nineteen hundreds and ameliorated further by Kubicek. It involves the recording of the instantaneous blood volume by the measurement of electrical impedance as a function of time. Therefore, the study of peripheral pulse morphology has gained much attention in the past few years among researchers. Physiological variability is one of the recent investigations added during the last two decades for the objective assessment of autonomic function and the assessment of prognosis in severe sicknesses namely myocardial infarction, diabetic neuropathy, etc. In addition to heart rate variability studied worldwide, few researchers have studied blood pressure variability and peripheral blood flow variability. In this computer era, artificial intelligence and machine learning techniques have become more important day-by-day, and different types of algorithms were used for the identification of hidden patterns from plethysmographic observations on the radial pulse such as support vector machine as well as crisp and fuzzy clustering. Eight patterns were classified with a yield of 80%–90% and helped with the diagnosis of disorders such as myocardial infarction, pulmonary tuberculosis, coronary artery disorders, cirrhosis of the liver, and bronchial asthma. This paper briefly describes the use of machine learning techniques for the classification of peripheral pulse morphologies.

Keywords

• crisp and fuzzy clustering
• impedance plethysmography
• machine learning
• peripheral pulse morphology
• support vector machine